Research map / Architectures and efficiency / Attention and transformers
Why Attention? Analyze BiLSTM Deficiency and Its Remedies in the Case of NER
Peng-Hsuan Li, Tsu-Jui Fu, Wei-Yun Ma
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Interleave forward and backward LSTM layers to model cross-context | Baseline-BiLSTM-CNN · OntoNotes 5.0 test | F1: 87.75±0.14 -> 88.27±0.17 | Improved |
| Add token-level self-attention over BiLSTM states | Baseline-BiLSTM-CNN · OntoNotes 5.0 test | F1: 87.75±0.14 -> 88.40±0.18 | Improved |
| Interleave forward and backward LSTM layers to model cross-context | Baseline-BiLSTM-CNN · WNUT 2017 test | F1: 40.68±1.78 -> 42.85±0.99 | Improved |
| Add token-level self-attention over BiLSTM states | Baseline-BiLSTM-CNN · WNUT 2017 test | F1: 40.68±1.78 -> 42.26±0.82 | Improved |
| Interleave forward and backward LSTM layers | Cross-BiLSTM-CNN · WNUT 2017 test | F1: The paper reports 2.2% improvement over baseline | Improved |
| Use cross-context modeling for three-token entity mentions | Cross-BiLSTM-CNN · WNUT 2017, mention length 3 | F1 improvement vs. Baseline: +8.7% | Improved |
| Use self-attention for three-token entity mentions | Att-BiLSTM-CNN · OntoNotes 5.0, mention length 3 | F1 improvement vs. Baseline: +2.3% | Improved |
| Zero token hidden states H at test time (ablation) | Att-BiLSTM-CNN · OntoNotes 5.0 dev, I tag | per-token recall: -28.18 | Hurt |
| Keep only context vectors C all at test time (ablation) | Att-BiLSTM-CNN · OntoNotes 5.0 dev, I tag | per-token recall: -3.80 | Hurt |
| Keep only attention head C2 at test time (ablation) | Att-BiLSTM-CNN · OntoNotes 5.0 dev, E tag | per-token recall: -36.45 | Hurt |
| Keep only attention head C3 at test time (ablation) | Att-BiLSTM-CNN · OntoNotes 5.0 dev, E tag | per-token recall: -39.19 | Hurt |
| Keep only attention head C4 at test time (ablation) | Att-BiLSTM-CNN · OntoNotes 5.0 dev, B tag | per-token recall: -12.21 | Hurt |
Closest papers by meaning
- ConsistNER: Towards Instructive NER Demonstrations for LLMs with the Consistency of Ontology and Context (AAAI 2024 · similarity 0.86)
- GuideNER: Annotation Guidelines Are Better than Examples for In-Context Named Entity Recognition (AAAI 2025 · similarity 0.86)
- MCL-NER: Cross-Lingual Named Entity Recognition via Multi-View Contrastive Learning (AAAI 2024 · similarity 0.84)
- Leveraging Attention to Effectively Compress Prompts for Long-Context LLMs (AAAI 2025 · similarity 0.84)
- A Systematic Exploration of Knowledge Graph Alignment with Large Language Models in Retrieval Augmented Generation (AAAI 2025 · similarity 0.84)
- NoLiMa: Long-Context Evaluation Beyond Literal Matching (ICML 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
